[PATCH v3 6/7] firmware: arm_ffa: Setup in-kernel users of FFA partitions

[Jens Wiklander]

On Wed, Jan 13, 2021 at 6:20 PM Sudeep Holla <sudeep.holla at arm.com> wrote:
>
> On Wed, Jan 13, 2021 at 01:58:56PM +0000, Achin Gupta wrote:
> > On Wed, Jan 13, 2021 at 01:30:56PM +0100, Jens Wiklander wrote:
> > > On Wed, Jan 13, 2021 at 10:44 AM Sudeep Holla <sudeep.holla at arm.com> wrote:
> > > [...]
> > > > > > > +static int ffa_partition_probe(const char *uuid_str,
> > > > > > > +                              struct ffa_partition_info *buffer)
> > > > > > > +{
> > > > > > > +       int count;
> > > > > > > +       uuid_t uuid;
> > > > > > > +       u32 uuid0_4[4] = { 0 };
> > > > > > > +
> > > > > > > +       if (uuid_parse(uuid_str, &uuid)) {
> > > > > > > +               pr_err("invalid uuid (%s)\n", uuid_str);
> > > > > > > +               return -ENODEV;
> > > > > > > +       }
> > > > > > > +
> > > > > > > +       export_uuid((u8 *)uuid0_4, &uuid);
> > > > > > > +       count = __ffa_partition_info_get(uuid0_4[0], uuid0_4[1], uuid0_4[2],
> > > > > > > +                                        uuid0_4[3], &buffer);
> > > > > Wrong byte order?
> > > > > According to section 5.3 of the SMCCC, UUIDs are returned as a single
> > > > > 128-bit value using the SMC32 calling convention. This value is mapped
> > > > > to argument registers x0-x3 on AArch64 (resp. r0-r3 on AArch32). x0
> > > > > for example shall hold bytes 0 to 3, with byte 0 in the low-order
> > > > > bits.
> > > > >
> > > >
> > > > I need to spend some time to understand the concern here. Initially I agreed
> > > > with your analysis and then a quick review make be realise it is all OK.
> > > > I need to check if my understanding is correct again. I thought I will
> > > > take example and check here itself.
> > > >
> > > > UUID: "fd02c9da-306c-48c7-a49c-bbd827ae86ee"
> >
> > IIUC this maps to (as per RFC4122).
> >
> > fd02c9da = time_low (bytes 0-3)
> > 306c48c7 = time_mid & time_hi_and_version (bytes 4-7)
> > a49cbbd8 = clock_seq_hi_and_reserved, clock_seq_low and bytes/octets 0-1 of node (bytes 8-11)
> > 27ae86ee = bytes 2-5 of node (bytes 12-15)
> >
> > SMCCC says:
> >
> > w0 : bytes 0-3 with byte 0 in the lower order bits.
> > w1 : bytes 4-7 with byte 4 in the lower order bits.
> > w2 : bytes 8-11 with byte 8 in the lower order bits.
> > w3 : bytes 12-15 with byte 12 in the lower order bits.
> >
> > This should amount to:
> >
> > w0 = dac902fd
> > w1 = c7486c30
> > w2 = d8bb9ca4
> > w3 = ee86ae27
> >
> > So, even though RFC4122 uses big-endian i.e network byte order. The UUID is
> > encoded as little-endian as per the SMCCC.
> >
> > What do you reckon?
> >
>
> Thank Achin, that matches my understanding too. I spent some time looking
> at RFC4122[1] and concluded what we have is fine.

Thanks for the analysis.

>
> @Jens, one thing to note, I am not claiming to support this driver with
> big-endian kernel. I plan to take that up once we settle with basic support.

No worries, the OP-TEE driver doesn't support that either.

Cheers,
Jens

>
> > cheers,
> > Achin
> >
> > > >
> > > > UUID[0]   UUID[1]  UUID[2]  UUID[3] (referring uuid0_4 above)
> > > > dac902fd c7486c30 d8bb9ca4 ee86ae27
> > > >
>
> Matches w0-w3 above, thanks for detailed explanation
>
> > > > It seems correct as per SMCCC convention to me, or am I missing something
> > > > obvious ?
> > >
> > > In this example I'd expect the first register to hold 0xfd02c9da
> > > regardless of the byte order of the machine. If there is a different
> > > byte order in the receiver it will still be received as 0xfd02c9da.
> > > That's how I've understood the specification.
> >
>
> --
> Regards,
> Sudeep
>
> [1] https://tools.ietf.org/html/rfc4122